We actively develop new methodology and explore learning theory relevant to applications in medical imaging. Areas of research include meta-learning, multi-task & continual learning, causality, domain shift, geometric deep learning, semi-supervised and unsupervised learning, representation learning and Bayesian methods.
Representation Learning
Morpho-MNIST: Quantitative Assessment and Diagnostics for Representation Learning (JMLR 2019)
Causality in Imaging
Causality matters in medical imaging (Nature Communications 2020)
Deep Structural Causal Models for Tractable Counterfactual Inference (NeurIPS 2020)
Machine Learning on Graphs
- Graph Convolutional Gaussian Processes (ICML 2019)
- Controlling Meshes via Curvature (IPMI 2019)
- Disease Prediction using Graph Convolutional Networks (MedIA 2018)
- Metric learning with spectral graph convolutions on brain connectivity networks (Neuroimage 2018)
Overfitting & Class-Imbalance
Overfitting of neural nets under class imbalance: Analysis and improvements for segmentation (MICCAI 2019)
Bayesian Deep Learning & Uncertainty Estimation
Implicit Weight Uncertainty in Neural Networks
Stochastic Segmentation Networks: Modelling Spatially Correlated Aleatoric Uncertainty (NeurIPS 2020)
Domain Shift
Domain Generalization via Model-Agnostic Learning of Semantic Features (NeurIPS 2019)
Unsupervised Domain Adaptation (IPMI 2017) and PnP-AdaNet (IEEE Access 2019)
Semi-Supervised Learning
Semi-supervised learning via compact latent space clustering (ICML 2018)
Multi-Task & Continual Learning
Towards continual learning in medical imaging (NeurIPS Workshop 2018).
